2-D classification question


I had a quick question about 2D classification. Many of my classes have the same number of particles regardless of how I classify them, which is preventing them from being distinct classes. Are there tips to help with this? Should I increase/decrease the number of particles I’m using? Increase/decrease the number of classes? I’ve seen this result with 20K to 200K particles, trying anywhere from 10 to 100 averages. Or is this a sample issue?


Hi @ebillin,

This is an interesting problem! Just to confirm, is it that you are seeing many 2D classes that represent essentially the same view/orientation of the particle, each containing around the same number of particles? If this is the case, it sounds like the sample may have strong preferred orientation. In this case, it might actually help to remove some of the over-represented views from the dataset, by running a Select 2D Classes job and intentionally leaving out some of the “duplicate” classes, then doing another round of 2D Classification. Another thing that you can try within 2D classification is to increase the “Initial classification uncertainty factor” parameter, from its default value of 2 to something between 4-10, which could help increase the diversity of classes that are produced.

One other thing that might be helpful to see how impactful the problem is, would be to run an Ab-initio Reconstruction job and examine the orientation distribution plots to see if preferred orientation is strongly present in the dataset. Signs of preferred orientation can also be seen in the output volume: for example, along the direction of preferred orientation, the volume might appear stretched or streaky. Generally, a mild amount of preferred orientation is fine, and removing over-represented views in the particle set is only necessary if the outputs of a reconstruction or refinement look biased or otherwise incorrect.

Also, could I ask which version of cryoSPARC you are using?


Thanks for getting back to me! Thanks for these suggestions. Yes most classifications are essentially the same orientation, so that may be an issue. I am using version 2.15.0.